CN1776743A - Texture classifying method based on phase coincidence - Google Patents
Texture classifying method based on phase coincidence Download PDFInfo
- Publication number
- CN1776743A CN1776743A CN 200510110660 CN200510110660A CN1776743A CN 1776743 A CN1776743 A CN 1776743A CN 200510110660 CN200510110660 CN 200510110660 CN 200510110660 A CN200510110660 A CN 200510110660A CN 1776743 A CN1776743 A CN 1776743A
- Authority
- CN
- China
- Prior art keywords
- image
- phase place
- texture
- image block
- method based
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The method includes steps: carrying out convolution between image block selected from original image and 2D log-Garbor filter so as to obtain phase coincidence image; restraining maximum and using double threshold method obtains phase coincidence mapped image i.e. texture edge image; then, calculating Shannon entropy of co-occurrence matrix of texture edge; finally, carrying out classification by using support vector machine and classifier. Characteristic point can be considered as a point with maximum phase coincidence among each harmonic wave of Fourier decomposition of signal, and the characteristic point possesses invariability of darkness/brightness of light beam and contrast. Advantages are: high precision of classification and good robustness.
Description
Technical field
The present invention relates to a kind of method of technical field of image processing, specifically is a kind of texture classifying method based on the phase place unanimity.
Background technology
Texture is a kind of image local feature comprehensive of visually-perceptible, is the pattern that gray scale or color produce with certain variation in the space.Existing analyzing image texture method is a lot, mainly contains statistical method, based on methods such as the method for model and signal Processing.Existing texture analysis method all is based on the amplitude information of image, i.e. the gray scale of image and color, and be subjected to the influence of environmental factor such as illuminance and contrast etc., in case these conditions change, the texture of image will be affected, thereby influences classification results.
Find by prior art documents, Liu etc. are at " IEEE Trans.on ImageProcessing " (" IEEE Flame Image Process journal ") June 2003,12 (6): " the Textureclassification using spectral histograms " that delivers on the 661-670 (" based on the histogrammic Texture classification of spectrum ") proposed the local feature that obtains based on image filtering and composed histogram, carries out the method for Texture classification.The distance of two spectrums between the histogram is by x
2-statistical measurement produces, and distance is little between the identical texture, and distance is big between the different textures.The notion of phase place unanimity can explain that square-wave signal can be a series of sine waves by fourier decomposition with the decomposition of square-wave signal, and at the rising edge and the negative edge of square wave, all sine waves have identical phase place, are called the phase place unanimity.This phenomenon is present in the peak value and the lowest point of triangular wave too, and the center of impulse function, and these points can signal unique point.The phase place unanimity is further improved through Kovesi (1993), expands to two dimension (2D) image by the two-dimentional log-Gabor wave filter of different scale and out of phase.
Summary of the invention
The objective of the invention is at the deficiencies in the prior art, a kind of texture classifying method based on the phase place unanimity is provided, its phase information with image is combined with support vector machine carry out Texture classification, the influence that nicety of grading is not changed by image light light and shade and contrast.
The present invention is achieved by the following technical solutions, and the present invention chooses image block earlier from original image, with two-dimentional log-Garbor wave filter convolution, obtain the initial phase coherent image; Suppress and the dual threshold method by maximum value again, obtain phase place consistent mapping graph, i.e. the texture outline map of image; Calculate the Shannon entropy of texture edge co-occurrence matrix then; Use support vector machine (SVM) to classify at last as sorter.
Describedly choose image block, be meant: all images that will classify are selected onesize image block, take all factors into consideration nicety of grading and calculate consuming timely, select the suitably image block of size from original image.In order to improve the classification degree of accuracy, selected image block should comprise texture information as much as possible, promptly image block should select as far as possible bigger; And the image block increase can cause that calculated amount is big, calculates consuming time manyly, thereby should reduce the image block size as far as possible.The present invention selects the image block of 64 * 64 pixels or 128 * 128 pixels as experimental image, can obtain better effects from nicety of grading and the compromise of calculating two aspects consuming time.
Described and two-dimentional log-Garbor wave filter convolution obtains the initial phase coherent image, is meant: original image is in yardstick s and orientation o and two-dimentional log-Garbor wave filter convolution, and the response amplitude that obtains is made as A
So(x, y), the phasing degree is φ
So(x y), utilizes phase deviation to estimate Δ φ
So(x y), and deducts estimating noise T from local energy, the phase place unanimity can be expressed as:
Wherein
Define besieged amount, when its value on the occasion of the time remain unchanged, when its value is negative value, then become 0; W (x) is the weights function of phase place unanimity.Original image piece and formula (1) convolution obtains the initial phase coherent image.
Described by maximum value inhibition and dual threshold method, obtain the consistent mapping graph of phase place, be specially: the initial phase coherent image is carried out bilinear interpolation, estimate on all directions of each pixel whether it is local maximum.Set two threshold values and come detection feature point, upper threshold value T
1With lower threshold value T
2, and T
1>T
2, and with all greater than T
1Point all be labeled as unique point.The present invention adopts 8 to face the territory, with all are labeled as the point of unique point and greater than T around the pixel
2Point all be labeled as unique point, and they are coupled together, form Feature Mapping figure.The Feature Mapping figure of this moment can be understood as the texture outline map of image.
The Shannon entropy of described edge calculation co-occurrence matrix is meant: to the texture outline map of image, displacement calculating is d=1 and 2 (for θ=0 °, 90 °) respectively, and
With
Edge co-occurrence matrix when (for θ=45 °) calculates its Shannon entropy then respectively, then has the input of six values as support vector machine.
Describedly classify as sorter, be meant:, classify as the input of support vector machine (SVM) with the proper vector that six Shannon entropys are formed with support vector machine.SVM adopts the polynomial kernel function, carries out classification experiments with the leave-one-out method.
The present invention is based on the analysis to image phase, and the each harmonic that unique point wherein can be understood as the fourier decomposition of signal has the point of maximum phase unanimity, and has the unchangeability to bright and dark light and contrast.So the present invention can obtain higher nicety of grading, the influence that not changed by bright and dark light and contrast.Owing to combine consistent characteristic of phase place and edge co-occurrence matrix, classify by support vector machine, have the advantage of good robustness.
Description of drawings
The experimental image that adopts in Fig. 1 embodiment of the invention
Embodiment
Below in conjunction with specific embodiment technical scheme of the present invention is described in further detail.
The experimental image that the embodiment of the invention adopts adopts the natural image of Brodatz natural texture collection.Whole invention implementation procedure is as follows:
1. from original image, choose earlier the image block of suitable size.Four image: D19 (wool spinning cloth) wherein, D29 (seabeach sand), the image block of D68 (grain of wood) and D84 (fiber of raffia leaf) is respectively as Fig. 1. (a), and (b), (c) and (d) (clear for image, as to be example) with the image block of 128 * 128 pixels.In the classification experiments, the image block of selecting 256 * 256,128 * 128,64 * 64,32 * 32 pixels and 16 * 16 pixels respectively is as experimental image.
2. calculate the initial phase coherent image.With original image piece I (x, y) carry out convolution by following transport function and 2Dlog-Gabor bank of filters:
Wherein, ω
0Be the centre frequency of wave filter, k/ ω
0For different ω
0It must be a constant.
Definition M
So EvenAnd M
So OddBe respectively even symmetry and odd symmetry wave filter, at yardstick s and orientation o, response vector can be obtained by image and wave filter convolution:
The amplitude of response is:
The phasing degree is:
φ
so(x,y)=atan(o
so(x,y)/e
so(x,y)) (5)
Definition φ
o(x is in the phasing degree of phase place o average y), and phase deviation is estimated and can be obtained by following formula:
ΔΦ
so(x,y)=cos(φ
so(x,y)- φ
o(x,y))-|sin(φ
so(x,y)- φ
o(x,y))| (6)
Simultaneously, consider the influence of noise, estimating noise T is deducted from local energy can eliminate the puppet response that noise produces the phase place unanimity, and the position of sharpening feature, so the phase place unanimity can be expressed as:
Wherein
Define besieged amount, when its value on the occasion of the time remain unchanged, when its value is negative value, then be 0; W (x) is the weights function of phase place unanimity.
Original image piece and formula (7) are carried out convolution obtain the initial phase coherent image, as Fig. 1. (e), (f), (g) and (h).
3. to the initial phase coherent image, suppress and the dual threshold refinement, produce the phase coincident characteristic mapping graph through maximum value.Non-maximum value suppresses to refer to the initial phase coherent image is carried out bilinear interpolation, estimates on all directions of each pixel whether it is local maximum.The dual threshold refinement is meant sets two threshold values, and T1 and T2, and T1>T2 are labeled as feature with all points greater than T1.Graphical analysis adopts 8 to face the territory, with all are labeled as the point of feature and value and all are labeled as unique point greater than the point of T2 around the pixel, couples together, and forms the phase coincident characteristic mapping graph, as Fig. 1. (i), (j), (k) with (l) shown in.At this moment, through the phase coincident characteristic mapping graph of refinement, can be understood as the texture outline map of image.
4. to the phase coincident characteristic mapping graph, quantize textural characteristics by the edge co-occurrence matrix that calculates texture.The definition of edge co-occurrence matrix derives from gray level co-occurrence matrixes, if apart (d, two gray-scale pixels θ) distribute to the Combined Frequency that occurs simultaneously, an available gray level co-occurrence matrixes H (d, θ) expression, wherein matrix element h in the image
IjNumerical value be all distances for d, with the horizontal direction angle be that the pixel of θ is to the number sum.If total K the gray level of image, then the size of gray level co-occurrence matrixes is K * K, and normalized gray level co-occurrence matrixes can be used C
0(i, j) expression.
In the texture outline map of image, gray-scale value has only 0 and 1, and the gray level co-occurrence matrixes of this moment just becomes 2 * 2 edge co-occurrence matrix, can be defined by the Shannon entropy:
The distance that the present invention adopts comprises d=1 and 2 (for θ=0 °, 90 °), and
With
(for θ=45 °) is so obtain six entropy.
5. the proper vector that six Shannon entropys of edge co-occurrence matrix are formed is classified as the input of support vector machine (SVM).SVM adopts the polynomial kernel function, carries out classification experiments with the leave-one-out method, and experimental result sees the following form.
Image size (pixel) | Classification accuracy rate (%) |
256×256 | 100 |
128×128 | 100 |
64×64 | 97.5 |
32×32 | 87.5 |
16×16 | 51.3 |
As can be seen from the table, when experimental image greater than 64 * 64 the time, can both obtain good classifying quality.Thereby as the compromise consuming time with calculating to experimental precision, the image block of easily selecting 64 * 64 pixels or 128 * 128 pixels for use is as experimental image.
Claims (7)
1, a kind of texture classifying method based on the phase place unanimity is characterized in that, chooses image block earlier from original image, with two-dimentional log-Garbor wave filter convolution, obtains the initial phase coherent image; Suppress and the dual threshold method by maximum value again, obtain phase place consistent mapping graph, i.e. the texture outline map of image; Calculate the Shannon entropy of texture edge co-occurrence matrix then; Classify as sorter with support vector machine at last.
2, the texture classifying method based on the phase place unanimity according to claim 1, it is characterized in that, describedly choose image block from original image, be meant: all images that will classify are selected onesize image block, take all factors into consideration nicety of grading and calculate consuming time, select the size of image block, selected image block should comprise texture information as much as possible.
3, according to claim 1 or 2 described texture classifying methods based on the phase place unanimity, it is characterized in that, describedly choose image block from original image, the image block of selecting 64 * 64 pixels or 128 * 128 pixels is as experimental image.
4, the texture classifying method based on the phase place unanimity according to claim 1, it is characterized in that, described and two-dimentional log-Garbor wave filter convolution, obtain the initial phase coherent image, be meant: suppose original image in yardstick s and orientation o and two-dimentional log-Garbor wave filter convolution, the amplitude of the response that obtains is A
So(x, y), the phasing degree is φ
So(x y), utilizes a phase deviation to estimate Δ φ simultaneously
So(x y), and deducts estimating noise T from local energy, then phase place is consistent is expressed as:
5, the texture classifying method based on the phase place unanimity according to claim 1, it is characterized in that, described by maximum value inhibition and dual threshold method, obtain the consistent mapping graph of phase place, be specially: non-maximum value suppresses to refer to the initial phase coherent image is carried out bilinear interpolation, estimates on all directions of each pixel whether it is local maximum; And dual threshold is meant with two threshold values and comes detection feature, upper threshold value T
1With lower threshold value T
2, and T
1>T
2, all are greater than T
1Point all be labeled as unique point, adopt 8 to face the territory, with all are labeled as the point of unique point and value greater than T around the pixel
2Point all be labeled as feature, and couple together, form the consistent mapping graph of phase place, the consistent mapping graph of the phase place of this moment is interpreted as the texture edge image.
6, the texture classifying method based on the phase place unanimity according to claim 1, it is characterized in that, the Shannon entropy of described calculating texture edge co-occurrence matrix, be meant: to the consistent mapping graph of phase place, when θ=0 ° and 90 °, displacement calculating is d=1 and 2 respectively, and during θ=45 °, displacement is
And
The edge co-occurrence matrix, calculate its Shannon entropy then respectively.
7, the texture classifying method based on the phase place unanimity according to claim 1, it is characterized in that, describedly classify as sorter with support vector machine, be meant: the proper vector that six Shannon entropys are formed is classified as the input of support vector machine, support vector machine adopts the polynomial kernel function, carries out classification experiments with the leave-one-out method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200510110660 CN1776743A (en) | 2005-11-24 | 2005-11-24 | Texture classifying method based on phase coincidence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200510110660 CN1776743A (en) | 2005-11-24 | 2005-11-24 | Texture classifying method based on phase coincidence |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1776743A true CN1776743A (en) | 2006-05-24 |
Family
ID=36766215
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200510110660 Pending CN1776743A (en) | 2005-11-24 | 2005-11-24 | Texture classifying method based on phase coincidence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1776743A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101109731B (en) * | 2007-08-08 | 2010-12-01 | 哈尔滨工业大学 | Gabor translating self-adapting window width selecting method represented by ultrasound signal |
CN101751673B (en) * | 2009-12-24 | 2012-05-23 | 中国资源卫星应用中心 | Multi-spectral image registration detection and correction method based on phase coincident characteristic |
CN102812357A (en) * | 2009-11-26 | 2012-12-05 | 马来西亚理工大学 | Methods And System For Recognizing Wood Species |
CN104766088A (en) * | 2014-01-07 | 2015-07-08 | 北京三星通信技术研究有限公司 | System and method of detecting object in three-dimensional image |
CN111693261A (en) * | 2020-05-28 | 2020-09-22 | 国网河北省电力有限公司电力科学研究院 | High-voltage shunt reactor iron core loosening fault diagnosis method |
CN111899261A (en) * | 2020-08-31 | 2020-11-06 | 西北工业大学 | Underwater image quality real-time evaluation method |
-
2005
- 2005-11-24 CN CN 200510110660 patent/CN1776743A/en active Pending
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101109731B (en) * | 2007-08-08 | 2010-12-01 | 哈尔滨工业大学 | Gabor translating self-adapting window width selecting method represented by ultrasound signal |
CN102812357A (en) * | 2009-11-26 | 2012-12-05 | 马来西亚理工大学 | Methods And System For Recognizing Wood Species |
CN101751673B (en) * | 2009-12-24 | 2012-05-23 | 中国资源卫星应用中心 | Multi-spectral image registration detection and correction method based on phase coincident characteristic |
CN104766088A (en) * | 2014-01-07 | 2015-07-08 | 北京三星通信技术研究有限公司 | System and method of detecting object in three-dimensional image |
CN111693261A (en) * | 2020-05-28 | 2020-09-22 | 国网河北省电力有限公司电力科学研究院 | High-voltage shunt reactor iron core loosening fault diagnosis method |
CN111693261B (en) * | 2020-05-28 | 2022-03-15 | 国网河北省电力有限公司电力科学研究院 | High-voltage shunt reactor iron core loosening fault diagnosis method |
CN111899261A (en) * | 2020-08-31 | 2020-11-06 | 西北工业大学 | Underwater image quality real-time evaluation method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110264448B (en) | Insulator fault detection method based on machine vision | |
CN113592845A (en) | Defect detection method and device for battery coating and storage medium | |
WO2015096535A1 (en) | Method for correcting fragmentary or deformed quadrangular image | |
CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
CN104240204B (en) | Solar silicon wafer and battery piece counting method based on image processing | |
CN103530590A (en) | DPM (direct part mark) two-dimensional code recognition system | |
CN1776743A (en) | Texture classifying method based on phase coincidence | |
CN105447512A (en) | Coarse-fine optical surface defect detection method and coarse-fine optical surface defect detection device | |
CN105740829A (en) | Scanning line processing based automatic reading method for pointer instrument | |
CN104834931A (en) | Improved SIFT algorithm based on wavelet transformation | |
CN105335972A (en) | Warp knitting fabric defect detection method based on wavelet contourlet transformation and visual saliency | |
CN108681737A (en) | A kind of complex illumination hypograph feature extracting method | |
CN112233116A (en) | Concave-convex mark visual detection method based on neighborhood decision and gray level co-occurrence matrix description | |
CN105404868A (en) | Interaction platform based method for rapidly detecting text in complex background | |
CN104680536A (en) | Method for detecting SAR image change by utilizing improved non-local average algorithm | |
CN116664565A (en) | Hidden crack detection method and system for photovoltaic solar cell | |
CN110276747B (en) | Insulator fault detection and fault rating method based on image analysis | |
CN107862689A (en) | Leather surface substantially damaged automatic identifying method and computer-readable recording medium | |
CN113222904B (en) | Concrete pavement crack detection method for improving PoolNet network structure | |
CN110246139A (en) | Planktonic organism in-situ image ROI rapid extracting method based on dual threshold | |
CN114120098A (en) | SAR image lakeshore line detection method and system based on MRSF model | |
CN113628170A (en) | Laser line extraction method and system based on deep learning | |
CN1790374A (en) | Face recognition method based on template matching | |
CN115063679B (en) | Pavement quality assessment method based on deep learning | |
Wu et al. | Research on crack detection algorithm of asphalt pavement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |